Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A method for background removal from documents, comprising: obtaining an image of a document; performing a clustering operation on the image to obtain a plurality of image segments; performing, for each image segment, a foreground/background classification to determine whether the image segment comprises foreground, wherein performing the foreground/background classification comprises: selecting a plurality of random patches of pixels in the image segment, classifying each of the selected random patches as either foreground or background, and based on the classification of the selected random patches, classifying the image segment as either foreground or background; obtaining an augmented image by combining the image segments comprising foreground; and obtaining a background-treated image by cropping the image of the document, based on the foreground in the augmented image.
This invention relates to document image processing, specifically background removal from scanned or photographed documents. The problem addressed is the presence of unwanted background elements in document images, which can interfere with optical character recognition (OCR) or visual analysis. The solution involves a multi-step process to isolate and remove background regions while preserving the document's foreground content. The method begins by capturing an image of a document. The image is then segmented into multiple regions through a clustering operation, which groups similar pixels together. For each segment, a foreground/background classification is performed. This classification involves selecting random pixel patches within the segment and individually classifying each patch as either foreground (document content) or background (unwanted elements). The segment is then classified as foreground or background based on the majority or statistical distribution of the patch classifications. Segments identified as foreground are combined to form an augmented image, which highlights the document's content. Finally, the original document image is cropped based on the foreground regions in the augmented image, resulting in a background-treated image where unwanted background elements are removed or minimized. This approach improves document clarity and readability for further processing or analysis.
2. The method of claim 1 , further comprising converting the image of the document to Lab color space, wherein the clustering operation is performed using ab channels of the Lab color space.
This invention relates to document image processing, specifically improving the accuracy of text extraction from scanned or photographed documents. The problem addressed is the difficulty of distinguishing text from background noise or complex layouts in images, particularly when using color-based segmentation techniques. The method involves analyzing a document image to identify text regions by clustering pixels based on color similarity. The image is first converted to the Lab color space, which separates luminance (L) from chrominance (ab channels). The clustering operation is then performed using only the ab channels, which represent color information without brightness variations. This approach enhances text detection by focusing on color differences rather than brightness, which can be affected by lighting conditions or document quality. The clustering process groups similar-colored pixels, allowing the system to identify text regions by their distinct color properties compared to the background. By isolating the ab channels, the method reduces interference from shadows, gradients, or uneven lighting, improving segmentation accuracy. This technique is particularly useful for documents with complex backgrounds or mixed content, such as forms, receipts, or handwritten notes, where traditional grayscale or RGB-based methods may fail. The invention builds on prior color-based segmentation methods but introduces a more robust approach by leveraging the Lab color space's perceptual uniformity and separating luminance from chrominance for more reliable clustering. This ensures better text extraction in real-world document processing applications.
3. The method of claim 1 , wherein performing the clustering operation comprises generating k image segments for k clusters, and wherein k represents the number of major color components identified in a color histogram of the image of the document.
This invention relates to image processing techniques for document analysis, specifically improving clustering operations to enhance color-based segmentation. The problem addressed is the need for accurate and efficient segmentation of document images into distinct color regions, which is crucial for tasks like document digitization, optical character recognition (OCR), and color-based data extraction. Traditional clustering methods often struggle with determining the optimal number of clusters or fail to adapt to the unique color distribution of each document. The invention improves upon prior methods by dynamically determining the number of clusters (k) based on the major color components identified in a color histogram of the document image. The clustering operation generates k image segments, where each segment corresponds to one of the k dominant color clusters. This approach ensures that the segmentation process adapts to the specific color characteristics of the document, leading to more precise and meaningful segmentation results. The method leverages color histograms to identify the most significant color components, which are then used to define the clusters. This dynamic clustering avoids the limitations of fixed-cluster methods, which may either over-segment or under-segment the image. The resulting segments can be used for further processing, such as text extraction, background removal, or color-based classification. The invention is particularly useful in applications where document images contain multiple distinct color regions, such as forms, tables, or annotated documents.
4. The method of claim 1 , wherein the clustering operation is performed using a K-means algorithm.
This invention relates to data processing systems that perform clustering operations to group similar data points. The problem addressed is the need for efficient and accurate clustering of data in various applications, such as machine learning, data analysis, and pattern recognition. The invention describes a method for clustering data points using a K-means algorithm. K-means is an unsupervised machine learning technique that partitions data into K distinct clusters based on their feature similarities. The algorithm iteratively assigns data points to the nearest cluster centroid and recalculates the centroids until convergence or a stopping criterion is met. The method involves initializing cluster centroids, typically randomly or using a heuristic approach. Data points are then assigned to the nearest centroid based on a distance metric, such as Euclidean distance. The centroids are updated by recalculating the mean of all points assigned to each cluster. This process repeats until the centroids stabilize or a predefined number of iterations is reached. The invention may also include preprocessing steps, such as data normalization or dimensionality reduction, to improve clustering performance. Post-processing steps, such as evaluating cluster quality using metrics like silhouette score or inertia, may also be included. The method is applicable to various domains, including image segmentation, customer segmentation, anomaly detection, and recommendation systems. The use of K-means ensures scalability and computational efficiency, making it suitable for large datasets.
5. The method of claim 1 , wherein classifying each of the selected random patches comprises applying a support vector machine configured to perform a binary classification between foreground and background.
This invention relates to image processing, specifically to methods for classifying regions within an image to distinguish between foreground and background elements. The problem addressed is the need for efficient and accurate segmentation of images into meaningful regions, particularly in applications such as object detection, medical imaging, or autonomous systems where distinguishing foreground objects from background is critical. The method involves selecting multiple random patches from an input image. Each patch is a small, localized region of the image. These patches are then classified to determine whether they belong to the foreground (objects of interest) or the background (non-relevant areas). The classification is performed using a support vector machine (SVM), a supervised machine learning algorithm optimized for binary classification tasks. The SVM is trained to distinguish between foreground and background based on features extracted from the patches, such as pixel intensity, texture, or other image characteristics. By applying the SVM to each selected patch, the method generates a classification result for each region, effectively segmenting the image into foreground and background areas. This approach improves the accuracy and efficiency of image segmentation by leveraging machine learning to analyze localized regions rather than processing the entire image at once. The method is particularly useful in scenarios where computational resources are limited or real-time processing is required.
6. The method of claim 1 , wherein cropping the image of the document comprises: binarizing the image of the document to distinguish foreground and background pixels based on the augmented image, obtaining a histogram based on a number of foreground pixels in columns of the binarized image, identifying in the histogram, a region in which the number of foreground pixels is below a specified threshold, and removing the region from the image of the document to obtain the background-treated image.
This invention relates to document image processing, specifically improving the quality of scanned or photographed documents by removing unwanted background regions. The problem addressed is the presence of extraneous background elements in document images, which can interfere with optical character recognition (OCR) or visual analysis. The solution involves a method to automatically crop or remove these background regions from the document image. The method begins by binarizing the document image to separate foreground (document content) from background pixels. This binarization is performed on an augmented version of the image, which may include enhancements like contrast adjustment or noise reduction. Next, a histogram is generated based on the number of foreground pixels in each column of the binarized image. The histogram is analyzed to identify regions where the number of foreground pixels falls below a specified threshold, indicating likely background areas. These regions are then removed from the image, resulting in a cleaned document image with reduced or eliminated background interference. The method ensures that only relevant document content remains, improving subsequent processing steps like OCR or archival storage. The threshold for foreground pixel density can be adjusted based on document type or image quality to optimize results.
7. The method of claim 6 , wherein cropping the image of the document further comprises: obtaining, for pixels in columns of the background-treated image, color variances, obtaining derivatives of the color variances, obtaining baseline variances at corners of the background-treated image, determine a cropping border in the derivatives of the color variances, based on a deviation from the baseline variances, and crop the background-treated image by applying the cropping border to the background-treated image.
This invention relates to document image processing, specifically improving the accuracy of cropping document images to remove unwanted background regions. The problem addressed is the difficulty in automatically detecting and removing background areas from document images, particularly when the background has varying colors or patterns that interfere with traditional edge-detection methods. The method involves analyzing a background-treated image of a document to determine precise cropping borders. First, color variances are calculated for pixels in the columns of the image. These variances are then used to compute derivatives, which help identify transitions between the document content and the background. Baseline variances are also obtained at the corners of the image to establish reference values. A cropping border is determined by analyzing deviations in the derivatives of the color variances relative to the baseline variances. The background-treated image is then cropped by applying this border, ensuring that only the relevant document content is retained while minimizing background interference. This approach enhances the accuracy of document cropping by leveraging color variance analysis and derivative-based edge detection, making it particularly useful for documents with complex or non-uniform backgrounds.
8. A system for background removal from documents, the system comprising: a computer processor; a pixel clustering engine executing on the computer processor configured to perform a clustering operation on an image of a document to obtain a plurality of image segments; and a foreground/background segmentation engine executing on the computer processor configured to: perform, for each image segment, a foreground/background classification to determine whether the image segment comprises foreground, wherein performing the foreground/background classification comprises: selecting a plurality of random patches of pixels in the image segment, classifying each of the selected random patches as either foreground or background, and based on the classification of the selected random patches, classifying the image segment as either foreground or background, and obtain an augmented image by combining the image segments that comprise foreground, and a cropping engine executing on the computer processor configured to: obtain a background-treated image by cropping the image of the document, based on the foreground in the augmented image.
A system for automated background removal from document images addresses the challenge of isolating text or relevant content from distracting or irrelevant backgrounds. The system processes an input document image using a computer processor to extract and retain only the foreground content, such as text or graphics, while discarding the background. The system includes a pixel clustering engine that divides the document image into multiple segments by grouping similar pixels. A foreground/background segmentation engine then analyzes each segment by randomly selecting multiple small pixel patches within the segment. Each patch is classified as either foreground or background, and the segment is classified based on the majority of these patch classifications. Segments identified as foreground are combined to form an augmented image containing only the relevant content. Finally, a cropping engine processes the augmented image to generate a background-treated image by cropping the original document image to the boundaries of the foreground content. This ensures the output image contains only the desired foreground elements, improving readability and usability for further processing or archival purposes. The system automates background removal, reducing manual effort and improving efficiency in document digitization and analysis.
9. The system of claim 8 , further comprising a color space conversion engine executing on the computer processor configured to convert the image of the document to Lab color space, wherein the clustering operation is performed using ab channels of the Lab color space.
This invention relates to document processing systems that analyze and cluster image data to identify and extract relevant information. The system addresses the challenge of accurately detecting and categorizing text, graphics, and other elements in scanned or digital documents, particularly when dealing with variations in color, contrast, or noise. The system includes a computer processor that executes a clustering operation to group similar pixels or regions within an image of a document based on their color characteristics. The clustering operation is performed using the ab channels of the Lab color space, which provides a perceptually uniform representation of color, making it easier to distinguish between different elements in the document. The system may also include a preprocessing module that prepares the image for analysis by applying filters, normalization, or other enhancements to improve the accuracy of the clustering operation. The clustering results can then be used for further processing, such as optical character recognition (OCR), document layout analysis, or content extraction. By leveraging the Lab color space, the system improves the robustness of document analysis in diverse imaging conditions.
10. The system of claim 8 , wherein performing the clustering operation comprises generating k image segments for k clusters, and wherein k represents the number of major color components identified in a color histogram of the image of the document.
The system processes digital images of documents to enhance readability by analyzing and clustering color components. The system identifies major color components in the document image by generating a color histogram, which quantifies the distribution of colors in the image. The number of major color components (k) determines the number of clusters (k) used in a clustering operation. The clustering operation segments the image into k distinct regions, each corresponding to a major color component. This segmentation helps isolate text or other relevant content from background noise or non-text elements, improving document analysis and optical character recognition (OCR) accuracy. The system dynamically adjusts the clustering based on the color histogram, ensuring adaptability to different document types and lighting conditions. The clustering operation may involve techniques such as k-means or other unsupervised learning methods to group pixels with similar color properties. The resulting segmented image allows for further processing, such as text extraction or background removal, to enhance document readability and automation in document processing workflows.
11. The system of claim 8 , wherein cropping the image of the document comprises: binarizing the image of the document to distinguish foreground and background pixels based on the augmented image, obtaining a histogram based on a number of foreground pixels in columns of the binarized image, identifying in the histogram, a region in which the number of foreground pixels is below a specified threshold, and removing the region from the image of the document to obtain the background-treated image.
This invention relates to document image processing, specifically to systems that automatically crop document images to remove unwanted background regions. The problem addressed is the presence of extraneous background areas in scanned or photographed documents, which can interfere with further processing such as optical character recognition (OCR) or archival storage. The system enhances document images by first binarizing the image to separate foreground (document content) from background pixels. A histogram is then generated, counting foreground pixels across vertical columns of the binarized image. The system identifies regions in the histogram where the number of foreground pixels falls below a predefined threshold, indicating areas likely to be background. These regions are automatically removed, resulting in a cropped image with reduced or eliminated background clutter. The method ensures that only the relevant document content remains, improving subsequent processing accuracy and storage efficiency. The approach leverages pixel density analysis to dynamically determine cropping boundaries, adapting to varying document layouts and background conditions. This technique is particularly useful in automated document workflows where manual cropping is impractical.
12. The system of claim 11 , wherein cropping the image of the document further comprises: obtaining, for pixels in columns of the background-treated image, color variances, obtaining derivatives of the color variances, obtaining baseline variances at corners of the background-treated image, determine a cropping border in the derivatives of the color variances, based on a deviation from the baseline variances, and crop the background-treated image by applying the cropping border to the background-treated image.
This invention relates to document image processing, specifically improving the accuracy of cropping document images to remove unwanted background regions. The problem addressed is the difficulty in automatically detecting and removing background areas in scanned or photographed documents, which often contain noise, shadows, or uneven lighting that can interfere with subsequent optical character recognition (OCR) or other document analysis tasks. The system processes a document image by first treating the background to enhance contrast between the document content and the background. It then analyzes the image by calculating color variances for pixels in vertical columns of the treated image. Derivatives of these color variances are computed to identify transitions between the document content and the background. Baseline variances are measured at the corners of the image to establish reference values. A cropping border is determined by detecting significant deviations from the baseline variances in the derivative data, which indicates the edges of the document content. The system then crops the image by applying this border, effectively removing the background while preserving the document content. This approach improves cropping accuracy by leveraging statistical analysis of color variations and their derivatives, ensuring that the cropped image retains the entire document while minimizing background interference. The method is particularly useful in automated document processing workflows where manual cropping is impractical.
13. A method for background removal from documents, comprising: obtaining an image of a document; performing a clustering operation on the image to obtain a plurality of image segments; performing, for each image segment, a foreground/background classification to determine whether the image segment comprises foreground; obtaining an augmented image by combining the image segments comprising foreground; and obtaining a background-treated image by cropping the image of the document, based on the foreground in the augmented image, wherein cropping the image of the document comprises: obtaining, for pixels in columns of the background-treated image, color variances, obtaining derivatives of the color variances, obtaining baseline variances at corners of the background-treated image, determine a cropping border in the derivatives of the color variances, based on a deviation from the baseline variances, and crop the background-treated image by applying the cropping border to the background-treated image.
The invention relates to a method for removing background from document images to enhance readability and focus on the foreground content. The method addresses the challenge of automatically identifying and isolating the relevant document content from surrounding or embedded background elements, such as shadows, borders, or noise, which can interfere with optical character recognition (OCR) or visual analysis. The process begins by capturing an image of a document. The image undergoes a clustering operation to segment it into multiple distinct regions or segments. Each segment is then classified as either foreground (the document content) or background (unwanted elements). The classified foreground segments are combined to form an augmented image, which highlights the document's primary content. The method then refines this result by cropping the original document image based on the foreground regions identified in the augmented image. Cropping involves analyzing the color variances of pixels in each column of the background-treated image. Derivatives of these variances are computed, along with baseline variances at the image corners. A cropping border is determined by detecting deviations from the baseline variances in the derivative data. The final cropped image is generated by applying this border to the background-treated image, effectively removing unwanted background regions while preserving the document's foreground content. This approach ensures precise and automated background removal, improving document processing accuracy.
14. The method of claim 13 , further comprising converting the image of the document to Lab color space, wherein the clustering operation is performed using ab channels of the Lab color space.
The invention relates to document image processing, specifically improving the accuracy of text extraction from scanned or photographed documents. The problem addressed is the difficulty in distinguishing text from background noise or complex layouts in documents, particularly when using color-based segmentation techniques. The method involves analyzing a document image to identify text regions by clustering pixels based on color similarity. The image is first converted to the Lab color space, which separates luminance (L) from color information (a and b channels). The clustering operation is then performed using only the a and b channels, which represent color without brightness interference. This allows for more accurate grouping of text pixels, as variations in lighting or shading do not affect the color-based clustering. The method may also include preprocessing steps like noise reduction and edge detection to enhance text detection before clustering. The final output is a segmented image where text regions are isolated for further processing, such as optical character recognition (OCR). This approach improves text extraction accuracy in documents with complex backgrounds or mixed content.
15. The method of claim 13 , wherein performing the clustering operation comprises generating k image segments for k clusters, and wherein k represents the number of major color components identified in a color histogram of the image of the document.
This invention relates to document image processing, specifically improving color-based segmentation for document analysis. The problem addressed is accurately identifying and clustering major color components in document images to enhance subsequent processing tasks like optical character recognition (OCR) or layout analysis. Traditional methods often struggle with documents containing multiple distinct color regions, leading to poor segmentation and downstream errors. The method involves analyzing a document image to generate a color histogram, which quantifies the distribution of colors present. From this histogram, the system identifies the k most significant color components, where k represents the number of distinct color regions deemed important for processing. Using these k color components, the image is segmented into k clusters, each corresponding to one of the major color regions. This clustering operation ensures that each segment contains pixels predominantly of a single identified color, improving the accuracy of subsequent document analysis tasks. The approach leverages color histogram analysis to dynamically determine the optimal number of clusters (k) based on the document's actual color distribution, rather than relying on predefined thresholds. This adaptability enhances performance across diverse document types, including those with complex color layouts. The segmented image can then be used for further processing, such as text extraction or layout reconstruction, with improved reliability.
16. The method of claim 13 , wherein cropping the image of the document further comprises: binarizing the image of the document to distinguish foreground and background pixels based on the augmented image, obtaining a histogram based on a number of foreground pixels in columns of the binarized image, identifying in the histogram, a region in which the number of foreground pixels is below a specified threshold, and removing the region from the image of the document to obtain the background-treated image.
This invention relates to document image processing, specifically improving the quality of scanned or photographed documents by removing unwanted background elements. The method addresses the problem of background noise, shadows, or irrelevant objects in document images, which can interfere with optical character recognition (OCR) or visual analysis. The process involves binarizing the document image to separate foreground text or graphics from the background. A histogram is generated based on the number of foreground pixels in each column of the binarized image. The histogram is analyzed to identify regions where the number of foreground pixels falls below a specified threshold, indicating areas likely dominated by background noise. These regions are then removed from the image, resulting in a cleaner, background-treated version. This technique enhances document clarity by focusing on the relevant content, improving subsequent processing steps such as OCR or machine learning-based analysis. The method is particularly useful for documents with complex backgrounds, shadows, or partial occlusions, ensuring accurate and reliable extraction of text and graphical elements.
17. A system for background removal from documents, the system comprising: a computer processor; a pixel clustering engine executing on the computer processor configured to perform a clustering operation on an image of a document to obtain a plurality of image segments; and a foreground/background segmentation engine executing on the computer processor configured to: perform, for each image segment, a foreground/background classification to determine whether the image segment comprises foreground, and obtain an augmented image by combining the image segments that comprise foreground, and a cropping engine executing on the computer processor configured to: obtain a background-treated image by cropping the image of the document, based on the foreground in the augmented image, wherein cropping the image of the document further comprises: obtaining, for pixels in columns of the background-treated image, color variances, obtaining derivatives of the color variances, obtaining baseline variances at corners of the background-treated image, determine a cropping border in the derivatives of the color variances, based on a deviation from the baseline variances, and crop the background-treated image by applying the cropping border to the background-treated image.
The system is designed for automated background removal from document images, addressing challenges in isolating text or relevant content from distracting or irrelevant backgrounds. The system processes an input document image using a computer processor with three key components: a pixel clustering engine, a foreground/background segmentation engine, and a cropping engine. The pixel clustering engine divides the document image into multiple segments by performing a clustering operation on the pixel data. The segmentation engine then analyzes each segment to classify whether it contains foreground content (e.g., text, graphics) or background. The classified foreground segments are combined into an augmented image, which serves as a reference for subsequent processing. The cropping engine refines the document image by removing background regions. This involves calculating color variances for each column of pixels in the image, computing derivatives of these variances, and determining baseline variances at the image corners. A cropping border is identified based on deviations from the baseline variances, and the image is cropped along this border to produce a final output with minimized background interference. The system automates background removal, improving document clarity and usability for further processing or analysis.
18. The system of claim 17 , further comprising a color space conversion engine executing on the computer processor configured to convert the image of the document to Lab color space, wherein the clustering operation is performed using ab channels of the Lab color space.
The invention relates to a document processing system that analyzes and processes document images to identify and extract relevant information. The system includes a computer processor and a memory storing instructions that, when executed, perform operations to process an image of a document. These operations include receiving the document image, performing a clustering operation on the image to identify clusters of pixels, and generating a representation of the document based on the clustering results. The clustering operation groups pixels with similar characteristics, such as color or intensity, to segment the document into distinct regions. The system further includes a color space conversion engine that converts the document image to the Lab color space, a color space that separates luminance (L) from chrominance (a and b channels). The clustering operation specifically uses the a and b channels of the Lab color space to analyze color differences while ignoring brightness variations, improving the accuracy of pixel grouping. This approach enhances document analysis by focusing on color distinctions, which is particularly useful for tasks like text extraction, background removal, or object detection in scanned or photographed documents. The system may also include additional components for further processing the clustered data, such as noise reduction or feature extraction.
19. The system of claim 17 , wherein performing the clustering operation comprises generating k image segments for k clusters, and wherein k represents the number of major color components identified in a color histogram of the image of the document.
This invention relates to document image processing, specifically a system for analyzing and segmenting document images based on color components. The system addresses the challenge of accurately extracting and organizing visual information from documents by leveraging color-based clustering techniques. The core functionality involves processing an image of a document to identify major color components through a color histogram analysis. The system then performs a clustering operation to generate k image segments, where k corresponds to the number of major color components detected. Each segment represents a distinct color region within the document, enabling precise separation of text, graphics, or other elements based on their dominant colors. This approach improves document digitization, optical character recognition (OCR), and automated content analysis by isolating relevant visual features for further processing. The clustering method ensures that the segmentation adapts dynamically to the document's color distribution, enhancing accuracy in diverse document types. The system may also include preprocessing steps to enhance image quality before clustering, such as noise reduction or contrast adjustment, to improve the reliability of color component identification. The output segments can be used for tasks like selective text extraction, background removal, or layout analysis, making the system valuable for applications in document management, archival systems, and automated data extraction workflows.
20. The system of claim 17 , wherein cropping the image of the document further comprises: binarizing the image of the document to distinguish foreground and background pixels based on the augmented image, obtaining a histogram based on a number of foreground pixels in columns of the binarized image, identifying in the histogram, a region in which the number of foreground pixels is below a specified threshold, and removing the region from the image of the document to obtain the background-treated image.
This invention relates to document image processing, specifically improving the quality of scanned or photographed documents by removing unwanted background elements. The system enhances document images by first binarizing the image to separate foreground (document content) from background pixels. A histogram is then generated based on the number of foreground pixels in each column of the binarized image. The system identifies regions in the histogram where the number of foreground pixels falls below a specified threshold, indicating areas likely to be background noise or irrelevant content. These regions are removed from the image, resulting in a cleaner, background-treated document image. This process helps eliminate distractions such as shadows, borders, or other non-document elements, improving readability and further processing accuracy. The method ensures that only the relevant document content remains, enhancing subsequent operations like optical character recognition (OCR) or archival storage. The system may also include additional preprocessing steps, such as skew correction or contrast adjustment, to further refine the document image before binarization and background removal. The invention is particularly useful in applications where document images are captured under varying conditions, such as mobile scanning or multi-page document digitization.
Unknown
August 11, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.